See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228367757

Automatic user support for business process modeling

Article

CITATIONS READS
37 953

4 authors:

Stefanie Betz Stefan Klink

Furtwangen University Duale Hochschule Baden-Württemberg Karlsruhe
79 PUBLICATIONS 1,084 CITATIONS 44 PUBLICATIONS 571 CITATIONS

SEE PROFILE SEE PROFILE

Agnes Koschmider Andreas Oberweis

University of Bayreuth Karlsruhe Institute of Technology
129 PUBLICATIONS 2,276 CITATIONS 229 PUBLICATIONS 3,086 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

A Crowdsourcing-based approach for individual e-exams View project

Privacy Framework for Process Mining View project

All content following this page was uploaded by Andreas Oberweis on 20 May 2014.

The user has requested enhancement of the downloaded file.

 Automatic User Support
for Business Process Modeling

Stefanie Betz, Stefan Klink, Agnes Koschmider, and Andreas Oberweis

Institute of Applied Informatics and Formal Description Methods

Universität Karlsruhe (TH), Germany
{betz,klink,koschmider,oberweis}@aifb.uni-karlsruhe.de

Abstract. The main purpose of business process modeling is the repre-

sentation and analysis of alternative process designs by formal or semi-
formal process models. Manual modeling of business processes is a time-
consuming task. Typos and structural modeling errors make it particu-
larly error prone to model business processes manually. Users can be
assisted in modeling business processes by providing an autocompletion
mechanism during the modeling process. In this paper we will describe
on-going work for autocompletion of business process models. This app-
roach is based upon an OWL DL description of Petri nets. Our au-
tocompletion mechanism requires validation methods to check process
properties of the automatically completed business process, which we
will introduce as well. Consequently, we aim to improve manual process
modeling by automating process modeling to a significant extent.

1 Introduction
The main purpose of business process modeling is the representation and analy-
sis of alternative process designs by formal or semiformal process models. Many
modeling languages – most of them being based on textual programming lan-
guages or graphical notations such as Petri nets [20], EPCs [23] or BPMN [27] –
have been proposed for process modeling. Novel orchestration- and choreography
languages such as BPEL [1] focus on tracking and executing business processes
by business applications. To enable verification of BPEL [8] proposes a Petri
net semantics. Petri nets have been established as a suitable language for mo-
deling business processes with intuitive graphical notation. Furthermore, Petri
nets have a mathematical foundation, which enables simulation and analysis of
system behavior.
In this paper we will describe on-going work for autocompletion of Petri
net based business process models. Manual modeling of business processes is a
time-consuming task. Typos and structural modeling errors make it particularly
error prone to model business processes manually. Users can be assisted in mo-
deling business processes by providing an autocompletion function during the
modeling process. However, process element names might differ in syntax even
when they have the same meaning (homonyms) or one process can be modeled
in different ways even when utilizing the same modeling language. It is possible
that these process elements will not be suggested as fitting elements to provide
autocompletion.
To solve ambiguity issues caused by the use of different names for descri-
bing the same tasks a machine readable and interpretable format, which might
be used for machine reasoning, is required for Petri nets. Business processes
modeled with Petri nets can be translated into the Web Ontology Language
OWL [17], an unambiguous format which allows ontological reasoning. So-called
semantic business process models combine process modeling methods with se-
mantic technologies to achieve automatic processing of business process models
instead of manual processing. We will use a semantic description of Petri nets to
make it easier to find appropriate process templates (reference processes), which
can be proposed for autocompletion. During the modeling process, a recommen-
dation mechanism determines possible subsequent fragments of all templates by
computing similarities. If the system detects a high similarity between one ele-
ment of a template and a modeling element, then subsequent elements of this
element template are proposed for autocompletion. To ensure correct process
flow behavior the system has additionally to check properties such as deadlock-
freeness.
The structure of this paper is as follows. Firstly, we will recall the main no-
tions of Petri nets and a semantic description of Petri nets with OWL DL. In
Section 3 we will describe an approach for measuring similarity between semantic
business process models by utilizing syntactic-, linguistic- and structural simi-
larity measurements. To validate process behavior properties we will illustrate
analysis methods in Section 4. Modeler’s behavior can be observed and learned
with machine learning techniques, which we will briefly survey in Section 5.
Section 6 concludes the paper with an outlook on future research.

2 Foundations

Next subsection introduces Petri net notation and Petri net modeling of business
processes.

2.1 Petri nets

Petri nets are a graphical language and a formalism used to model business
processes and verify system behavior. Formally, a Petri net is a directed bipartite
graph with nodes and arcs. It can be described by the triple N = (P, T, F ),
where P is a set of places, T is a set of transitions (which is disjoint from P )
and F ⊆ (P × T ) ∪ (T × P ) is a flow relation. Elements of P are graphically
represented as circles, elements of T as boxes and elements of F as directed
arcs between places and transitions. A place p is an input place of a transition
t, if there exists a directed arc from p to t. A place p is an output place of a
transition t, if there exists a directed arc from t to p. The set of all input places
of a transition t is denoted by •t and is called preset. The set of all output places
is denoted by t• and is called postset.
 Numerous Petri net variants have been proposed, which can be subsumed
in elementary or high-level Petri nets. In elementary Petri nets places contain
tokens, which represent anonymous objects. A transition t is enabled to change
the marking m of the net, if each place p in the preset contains at least one
token. If such a transition t occurs, then t consumes tokens from the preset and
inserts tokens in the postset. The occurrence sequence, which leads from m to
t
m0 is defined by m → m0 . A marking m0 is called reachable from m, if there
δ
exists an occurrence sequence from m to m0 (m → m0 ) with δ = t1 , t2 . . . tn .
For modeling system behavior or business processes with Petri nets we dis-
tinguish several process flow structures [26] as depicted in Figure 1. The flow
structure OR-split allows to model alternative branching. The two alternative
branches are again integrated by a so-called OR-join or a so-called AND-join
(synchronization). By the use of AND-split (concurrency) tokens are distributed
to two places.

Fig. 1. Overview of Process Flow Structures

The formal foundation of Petri nets allows to verify whether a modeled busi-
ness process meets certain properties such as deadlock-freeness. If a marking is
reached, which is not an intended final process state and which doesn’t enable
any transition, then the process is in a deadlock, as depicted in Figure 2.

Fig. 2. Deadlock in a Petri net

To make tokens distinguishable, variants of high-level Petri nets have been

proposed such as Coloured Petri nets [6] or Predicate/Transitions nets [10]. In
this paper, we focus on Predicate/Transitions nets (Pr/T nets) where places
are interpreted as predicates representing relation schemes. Transitions occur
according to a logical expression which may be attached to them.
2.2 Semantic Business Process Models

To describe business processes modeled with Petri nets in an unambiguous for-

mat we have proposed a translation of Pr/T nets to OWL DL [13]. For our work
we will refer to OWL DL (Description Logic) in order to be able to use available
off-the-shelf reasoning technologies. OWL DL can be considered as a syntactic
variant of the SHOIN (D) Description Logic which is known to be decidable
[9]. In Table 1 the main constructs of SHOIN (D) are shown.

Table 1. Constructs of SHOIN (D) Syntax

subClass A1 v A2
intersection A1 u ... u An
union of A1 t ... t An
allValuesFrom ∀ P.A
someValuesFrom ∃ P.A
maxCardinality ≤ nP
minCardinality ≥ nP

In our Pr/T net ontology each Petri net element corresponds to an OWL
concept. P laces are described by the concept Place, transitions by Transition
and arcs by F romP lace (P × T ) and T oP lace(T × P ). For instance, the concept
PetriNet is defined by at least one transition, place and arc.

Table 2. Pr/T net Ontology

P etriN et ≡ ≥ 1hasN ode.(T ransition u P lace)

u ≥ 1hasArc.(F romP lace t T oP lace)
T ransition ≡ placeRef.P laceu = 1haslogicalConcept.LogicalConcept
P lace ≡ transRef.T ransitionu = 1hasM arking.IndividualDataItem
F romP lace ≡ ≥ 1hasInscription.Delete u ∃hasN ode.P lace
T oP lace ≡ ≥ 1hasInscription.Insert u ∃hasN ode.T ransition
LogicalConcept ≡ = 1hasConditon.Conditiont = 1has.Operation.Operationu
∃hasAttribute.IndividualDataItem
IndividualDataItem ≡ ≥ 1hasAttribute.Attribute
Delete ≡ ∀hasAttribute.IndividualDataItem
Insert ≡ ∃hasAttribute.IndividualDataItem
Atrribute ≡ ≤ 1hasV alue.V alue
V alue ≡ hasRef.V alue
Condition ≡ f orall(string) t exists(string) t and(string)
Operation ≡ f unction(string)

To the best of our knowledge, there exists no other approach that transforms
high-level Petri nets into OWL DL. In the next section we will explain how to
measure similarity between a pair of semantic business process model elements.
3 Measuring Similarity between Process Elements

Before proposing appropriate process fragments, the recommendation mecha-

nism has to compare modeling elements with process templates and has to find
similar elements, which will be proposed as fitting subsequent elements. Thus,
the mechanism has to compare process templates with process elements, which
are currently modeled, by computing their similarities. The autocompletion sys-
tem observes and learns from the modeler’s behavior. Figure 3 gives an overview
of our recommendation process1 . Instead of representing OWL syntax, which is
not readable for modelers, we have depicted business processes in graphical Petri
net notation. However, the automatical similarity computation between process
elements is based upon the OWL serialisation of Petri nets.

Fig. 3. Overview of the Auto Completion process

Our approach for automatic process element propositions is similar to the

autocompletion function in mobile phones. The installed system observes what
the user is currently typing, and tries to complete words automatically.
We propose to compute similarities between two process element names by
utilizing the similarity measures syntactical, linguistic and structural measure-
ments. Syntactical similarity measures take as input two character strings and
compare them. A well-established string similarity measurement has been pro-
posed by Levenshtein [14] which takes into account the amount of operations
(insertion, deletion and substitution) needed to transform one string into an-
other. For example, to turn flight request to request requires seven deletions.
Based on the Levenshtein method [16] has proposed a syntactic similarity mea-
surement which returns for the similarity calculation similarity degrees between
0 and 1. But, syntactical similarity measurements alone are not sufficient since
they do not regard the semantics of words. In addition to the syntactical similar-
ity measurement we are utilizing background knowledge in terms of ontologies.
1
different shading of process templates visualizes independent process models.
 Petri nets obey an operational semantics that describes the control flow.
However, a missing semantic description of Petri net components hampers the
automated processing of process elements. To uncover synonyms or homonyms
of process element names we need a description of Petri nets in an umambiguous
format such as OWL DL. With the background ontology – numerous background
ontologies are modeled with OWL – we compute linguistic similarity measures.
Ontologies have paved the way for standardized formal conceptualizations of all
kinds of knowledge. To compute linguistic similarity degrees we have worked
with WordNet2 via the JWNL API [5] and with a specific UML Profile [4]. The
benefit of the UML-based background ontology is that a converting tool provides
an automatic translation from the visual UML modeling to OWL DL syntax.
WordNet is in contrast to the UML Profile predefined and fix.
However, to improve the aggregation of syntactical and linguistic similarity
measurements we take into account the context of element names by considering
structural information of names. For instance, the structure of place names is
influenced by places attributes, values and the subsequent transitions as depicted
in Figure 4 exemplarily for the place flight request. This place has a marking with
the name R flight request and the corresponding attributes Name, Destination,
Date and Quantity where Destination has the values PAR (corresponding to the
destination Paris) and FRA (corresponding to the destination Frankfurt). The
subsequent transition of flight request is check request.

Fig. 4. Context of the Place flight request

However, each concept influences place names with different weights. There-
fore, we have determined different weights for the concepts as shown in Table
3. In business process models with a lot of places, attributes and transitions
weights play a less important role than in small processes with less process ele-
ments. The more instances are modeled in a SBPM the more extensive is the
context of instances. The processes modeled in this paper have only few places
and transitions3 . Furthermore, depending on the features the similarity mea-
sure might differ. To determine structural similarity between place names we
compute the syntactical similarity degree of names (measure for names is syn-
2
an english online lexical reference system, which provides synonym and hypernym
sets consisting of nouns, verbs, adjectives, and adverbs [18].
3
To make the business processes readable we did not assign values for attributes
tactical similarity) and the linguistic similarity degrees of their attributes, values
and subsequent transitions.

Table 3. Features and Similarity Measures for Petri nets

Comparing Feature Measure Weight

Places name synt sim. 0.2
Attribute/Value str sim. 0.5
successor ling sim. 0.3
Attributes name synt sim. 0.2
sibling Attribute ling sim. 0.3
Values ling sim. 0.3
Place ling sim. 0.2
Values name synt sim. 0.2
Attribute ling sim. 0.5
Value reference ling sim. 0.3
Transitions name synt sim. 0.2
ToPlace sling sim. 0.4
FromPlace ling sim. 0.4

By aggregating these three similarity measures to a combined similarity mea-

sure we consider syntactical, linguistic, and structural properties of elements and
can compute a more significant similarity between two elements. Table 4 shows
some combined similarity degrees (simcom ) for process names (nameBP ) and
fragments (nameBF ). If the algorithm computes a combined similarity degree
> 0.5 between a template element and a process element, which is currently
modeled, then the recommendation system proposes it for autocompletion. The
user can then decide if (s)he accepts the proposition or rejects it. To learn a
threshold θ instead of using a fix value we will present in Section 5 several ma-
chine learning techniques.

Table 4. Results of Combined Similarity

nameBP nameBF simcom

flight request request 0.8
check request check request 1.0
request checked checked 0.9
... ... ...

In the following section we will sketch a method for validating behavior pro-
perties of autocompleted business processes.
4 Analysis Methods for Petri nets
To check if a process model meets certain properties, several analyzing methods
have been proposed [25]. In our approach we utilize such methods to validate
that the insertion of the proposed process fragments does not cause deadlocks
and synchronization errors. Thus, our automatic user support includes more
than the recommendation of process fragments.
In the following we are focusing on validation of process properties by check-
ing if the autocompleted business process is deadlock free and without lack of
synchronization. While utilizing validation algorithms for Pr/T nets we are not
considering any inscriptions of places, transitions, or arcs. Instead we are fo-
cusing on the net structure of the modeled process by identifying all possible
flows based on instance subgraphs [22] (see Figure 5 where we have modeled the
instance subgraph for Figure 2). We assume, that the analysis of a specific Pr/T
net satisfies the following requirements: the Pr/T net has only one source place
and one sink place, every place p and every transition t is on the path between
the source and the sink [24]. Additionally, cycles in the net structure must be
regarded as single execution units.

Fig. 5. Instance Subgraphs of Petri net in Figure 2

A deadlock can occur, if branches of an OR-split (such a split can be easier

identified by an instance subgraph) are synchronized by an AND-join, as depicted
in Figure 2. For instance, if the user inserts the transition complement data
and its subsequent place data complemented and intends to synchronize these
elements by inserting a connection from checked and data complemented to send
data, then a deadlock occurs.
A lack of synchronization occurs, if an AND-split is synchronized by an OR-
join, as depicted in Figure 6. For instance, if the user inserts the transition travel
request and intends to synchronize the concurrent branches by inserting the place
transport request, then a lack of synchronization occurs. To facilitate manual
modeling we aim to develop a recommendation mechanism, which advices only
validated fragments.
We have sketched how deadlocks and lack of synchronization can be discov-
ered and now we will introduce reduction rules to validate both process properties
[22], [15], [25]. According to the reduction rules a process is free of structural
 Fig. 6. Example for a Lack of Synchronization

errors if the reduction results in an empty graph. Because of the simplification of

the presented processes, we applied only two of the rules, the terminal reduction
rule (trr) and the sequential reduction rule (srr) as shown in Figure 7. At the
end there are two nodes left over, so the modeled process is not free of deadlocks.

Fig. 7. Execution of the Reduction

5 Learning User Behavior

As described in Section 3 our system is capable of recommending similar ele-
ments and the user is asked to select appropriate templates. For further recom-
mendations the users behavior could be observed, i.e., logging which decisions
have been made in which situation/context and at which process position or
template.
To recommend appropriate elements, a classification method must be utilized
to classify the set of stored elements into partitions. These partitions can then be
compared with the current user-behavior and the best one(s) concerning context
and other criteria as described above will be recommended for selection. Several
methods can be utilized to learn user’s behavior. The following list gives an
overview of established learning techniques.
Support Vector Machines The goal of a support vector machine (SVM) is
a classifier in a high-dimensional feature space which can handle even not-
separable problems. SVMs can be used to find hypotheses which guarantee
a minimal error [11]. SVMs determine those coefficients which separate the
training set with the shortest distance vectors. Only the so-called support
vectors are of interest and are spanning the separating hyperplane. One ad-
vantage of SVMs is that they can learn non-linear hypotheses by using a
 mapping function which maps the original feature space into a (commonly)
higher dimensional feature space or by using learning techniques like poly-
nom classifiers, radial-basis classifiers [3] or two-hidden layer (2HL) networks
which can form even more complex decision boundaries [2]. Another advan-
tage of SVMs is that they can handle noise-(error-)reduction by deleting
those examples – in our case the observed user-behavior – which cause the
non-separability. SVMs are robust and do not need any parameter optimiza-
tion, i.e., they can work on raw data.
Neural Nets Neural nets are a well-established and a successfully constituted
technique in the area of soft computing to process information in a vague
and tolerant manner. But the most important feature of neural nets is their
learning capability which increases the adaptivity of the system and supports
the complexity of heterogenous data. A variety of models are available to
choose. For recommender systems in general backpropagation algorithms
are especially suitable to handle the users feedback and selections to train
the net. In our case, neural nets can be used for example to learn a threshold
θ in Section 3 for the combined similarity – instead of using a threshold with
a fixed value of 0.5.
Bayesian Networks Alternative names are belief networks, probabilistic in-
dependence networks, influence diagrams, or causal nets [19]. A Bayesian
network is a directed acyclic graph (DAG) which represents probabilities of
dependencies between a set of random features [7]. The nodes of the graph
represent the set of random variables X = {X1 , . . . , Xn } and the directed
edges represent the dependency between these variables. The second part of
a Bayesian network is a set of conditional probabilities P (Xij |Xi ) according
to the associated graph. One advantage of Bayesian networks is that the
representation of a probability distribution as a directed graph enables the
analysis of complex conditional user events with a graph theoretical approach
and ensures the consistency of the system [21]. With Bayesian networks it
is possible to calculate conditional probabilities, i.e., to estimate the users
future behavior in case of observations made in the past.
Information Filtering Information Filtering is well known from shopping web
sites. When the customer clicks on an item I1 , then the system proposes:
”Customers who have bought this item I1 also have bought another item I2
too. . .”. The advantage of this technique is its simplicity. It can be imple-
mented very easily (only log files have to be parsed) and it can learn incre-
mentally. In our case, Information Filtering can be used to learn behaviors
like: ”Customers who have inserted this process template at this position
also have done this and that. . .”. Furthermore, with Information Filtering
techniques the system is able to generate and to compare user-profiles which
help to categorize users, e.g., in beginners or specialists [12].
Content-based Information Filtering Content-based Information Filtering
is a new extension which also takes the content of the underlying items
into account, i.e. I1 and I2 is also compared and their similarity is regarded
while ranking. With this method the combined similarity of process template
elements described in section 3 can be used to calculate the similarity and
 to rank the recommendations retrieved by the Content-based Information
Filtering technique.

Summarizing the upper list, especially each variant of Information Filtering

is a promising technique for recommending process templates, which will be
further considered.

6 Conclusion

In this paper we have presented on-going work for assisting users in business pro-
cesses modeling. Compared to manual process modeling we aim to improve the
reusability of business processes by an autocompletion mechanism. We propose
to use a recommendation system, which observes the user’s behavior and suggests
possible subsequent elements. Furthermore, the system facilitates validated pro-
cess properties of the automatically completed process to avoid deadlocks and
lacks of synchronization.
Modeling languages such as EPCs, BPMN or BPEL have no direct formal
foundations and thus do not enable analysis methods. Hence, if the process is
modeled with these languages the process to be automatically completed can
not be directly validated regarding deadlocks or lacks of synchronization.
Currently, we are developing an algorithm to automatically analyze hierar-
chical specifications of process elements and to detect errors in process models.
Processes with hierarchical specifications generally appear to be more intuitive
and easier to understand.

References

1. A. Arkin, S. Askary, B. Bloch, F. Curbera, Y. Goland, N. Kartha, C. K. Liu,

S. Thatte, P. Yendluri, and A. Yiu. Web services business process execution lan-
guage version 2.0. wsbpel-specification-draft-01, OASIS, September 2005.
2. R. Beale and T. Jackson. Neural Computing: An Introduction. Institute of Physics
Publishing, Bristol, U.K. and Philadelphia, PA, 1990.
3. C. M. Bishop. Neural networks for pattern recognition. Technical report, Oxford,
Clarendon, P., 1995.
4. S. Brockmans, M. Ehrig, A. Koschmider, A. Oberweis, and R. Studer. Semantic
Alignment of Business Processes. In Proceedings of the 8th International Confer-
ence on Enterprise Information Systems, Paphos, Cyprus, May 2006. to appear.
5. J. Didion. JWNL 1.3, November 2003. http://www.codezoo.com/pub/component/
196?/category=5.
6. H. J. Genrich and K. Lautenbach. System modelling with high level petri nets.
Theoretical Computer Science, (13):109–136, 1981.
7. D. Heckerman. A Tutorial on Learning with Bayesian Networks. Technical report
MSR-TR-95-06, Microsoft Research, March 1995.
8. S. Hinz, K. Schmidt, and C. Stahl. Transforming BPEL to Petri Nets. In Business
Process Management, pages 220–235, 2005.
 9. I. Horrocks. Applications of Description Logics: State of the Art and Research
Challenges. In Proceedings of the International Conference on Conceptual Struc-
tures, Lecture Notes in Computer Science, pages 78–90. Springer, 2005.
10. K. Jensen. An Introduction to the Theoretical Aspects of Coloured Petri Nets. In
J. de Bakker, W. P. de Roever, and G. Rozenberg, editors, A Decade of Concur-
rency – Reflections and Perspectives, Lecture Notes in Computer Science, pages
230–272. Springer, 1994.
11. T. Joachims. Text Categorization with Support Vector Machines: Learning with
Many Relevant Features. In C. Nedellec and C. Rouveirol, editors, Proceedings of
the 10th European Conference on Machine Learning (ECML’98), Lecture Notes in
Computer Science, pages 137–142, Chemnitz, Germany, 1998.
12. S. Klink. Query reformulation with collaborative concept-based expansion. In
Proceedings of the First International Workshop on Web Document Analysis (WDA
2001), pages 19–22, Seattle, Washington, USA, 2001.
13. A. Koschmider and A. Oberweis. Ontology based Business Process Description.
In J. Castro and E. Teniente, editors, Proceedings of the CAiSE-05 Workshops,
Lecture Notes in Computer Science, pages 321–333, Porto, Portugal, June 2005.
14. V. I. Levenshtein. Binary Code capable of correcting deletions, insertions and
reversals. Cybernetics and Control Theory, (8):707–710, 1966.
15. H. Lin, Z. Zhao, H. Li, and Z. Chen. A Novel Graph Reduction Algorithm to
Identify Structural Conflicts. In Proceedings of the 35th Hawaii international Con-
ference on System Sciences, pages 536–550. IEEE Computer Society Press, 2002.
16. A. Maedche and S. Staab. Measuring similarity between ontologies. In Proceedings
of the European Conference on Knowledge Acquisition and Management, Lecture
Notes in Computer Science, 2002.
17. D. L. McGuinness and F. van Harmelen. OWL Web Ontology Language Overview.
W3c recommendation, World Wide Web Consortium, 2004.
18. G. A. Miller, C. Fellbaum, and R. Tengi. WordNet – a lexical database for the
English language, 2006. http://wordnet.princeton.edu/.
19. J. Pearl. Probabilistic Reasoning in Intelligent Systems: Networks of Plausible
Inference. Morgan Kaufmann, San Mateo, 1988.
20. W. Reisig and G. Rozenberg. Lectures on Petri Nets: Basic Models. Lecture Notes
in Computer Science. Springer, 1 edition, 1998.
21. B. A. Ribeiro-Neto and R. Muntz. A belief network model for IR. In H.-P. Frei,
D. Harman, P. Schäuble, and R. Wilkinson, editors, Proceedings of the 19th Annual
International ACM SIGIR Conference on Research and Development in Informa-
tion Retrieval, pages 252–260, Zurich, Switzerland, August 18-22 1996. ACM Press.
22. W. Sadiq and M. E. Orlowska. Analyzing Process Models Using Graph Reduction
Techniques. Inf. Syst., (2):117–134, 2000.
23. A.-W. Scheer and M. Nüttgens. ARIS Architecture and Reference Models for Busi-
ness Process Management. In Business Process Management, Models, Techniques,
and Empirical Studies, volume 1806, pages 376–389. Springer, 2000.
24. W. M. P. van der Aalst. Workflow verification: Finding control-flow errors using
petri-net-based techniques. In Business Process Management, pages 161–183, 2000.
25. W. M. P. van der Aalst, A. Hirnschall, and H. M. W. E. Verbeek. An Alternative
Way to Analyze Workflow Graphs. In CAiSE, pages 535–552, 2002.
26. W. M. P. van der Aalst, A. H. M. ter Hofstede, B. Kiepuszewski, and A. P. Barros.
Workflow patterns. Distributed and Parallel Databases, 14(1):5–51, 2003.
27. S. A. White. Business Process Modeling Notation. Specification, BPMI.org, 2004.

View publication stats